车路协同环境下大范围路网运行状态群智感知与预测研究

As an internationally recognized revolutionary means to improve road traffic safety and efficiency, vehicle-to-infrastructure cooperation system has been applied to many situations. Due to the complexity, randomness and time-varying characteristics of traffic system, even the traffic elements are interconnected under vehicle-to-infrastructure cooperation system, the precise prediction and multi-scale evolution analysis of road network state is difficult to realize. Research is performed aiming at the characteristics of large scale road network state under vehicle-to-infrastructure cooperation condition. The incentive mechanism of large scale road network crowd sensing is proposed based on the autonomous decisions of the mobile sensors. The intelligent cooperative perception of large scale road network state under the vehicle-to-infrastructure cooperation could be achieved. By extracting road network spatial and temporal characteristics , the traffic state prediction model and multi-scale evolution model are constructed based on the deep learning and reinforcement learning method. This study could provide an important support for the realization of agile decision-making and optimized and coordinated control for large scale road network under the vehicle-to-infrastructure cooperation condition.

车路协同作为国际公认的提升道路交通安全与效率的革命性手段，正以场景为驱动逐步实现大范围应用。由于道路交通系统的复杂、随机、时变等特性，尽管车路协同环境下交通要素互联互通，但仍难以实现大范围路网全时全域的状态精准感知以及演化过程的多尺度分析。本项目针对车路协同环境下的大范围路网运行状态节点异质多样、场景复杂时变、数据价值密度高低并存等特点，以移动群体参与节点的自主决策为基础，提出基于博弈论的大范围路网群智感知计算激励机制，研究车路协同环境下的大范围路网运行状态智能协作感知方法，利用深度学习、强化学习网络表达能力、泛化能力强等特性，提取路网运行状态时空特征，针对常态与非常态情况分别构建深层次的路网状态预测与演化组合模型，为车路协同环境下的大范围路网实现敏捷决策和协同优化控制提供重要的支撑。

由于道路交通系统复杂、随机、时变等特性，导致系统时空高维的动态变化与感知、预测的精准和高效形成矛盾，全时全域状态的精准感知以及演化过程的多尺度分析成为实现道路交通协同控制的一个技术难题。为解决上述难题，本课题基于路段连通重要度分析和节点效用计算评估，构建车路协同环境下城市路网群智感知系统服务节点选择优化模型，并综合考虑系统收益提出了基于车辆信誉的激励机制，实现大范围路网运行状态的智能协作感知；在获取感知数据基础上，提取大范围路网状态时空特征，并采用二维矩阵表征路网状态，实现路网运行状态的数字化表征；然后基于深度学习和强化学习构建常态和非常态下路网状态预测模型，揭示大范围路网状态演化规律；最后，通过实际路网运行数据对所提模型进行验证，结果表明本课题提出的预测模型具有较高的预测精度，预测能力稳定可靠。课题组顺利完成任务书规定研究内容，累计发表SCI论文12篇、EI会议论文6篇，申请发明专利10项，获得软件著作权2项，出版学术专著1部。课题成果支撑北京路网运行态势演化分析与调控平台、车联网协同感知与控制技术装备等研发，为车路协同环境下的道路交通系统实现敏捷和优化、协同的控制和决策提供了重要支撑，在车路协同、智慧城市建设等领域具有广阔应用前景。